Computer vision based defect detection in hot-rolled steel surfaces in industrial manufacturing.
In cases when there are machine calibration issues, environmental settings, or equipment malfunction, the entire batch of production may become faulty. In such cases, manual inspection after the fact may prove to be expensive, as the items may have already been produced and the entire batch of faulty products (maybe hundreds or thousands) may need to be discarded.
In summary, the manual process of inspection is slow, inaccurate, and expensive.
A computer vision–based visual inspection system can detect surface defects in real time by analyzing streams of video frames. The system can send alerts, in real time, when a defect or a series of defects is detected so that the production can be stopped to avoid any loss.
We have developed a machine learning based model to detect visual defects in hot rolled steel surfaces. The system is able to detect 6 types of defects.